Parameters
26.2B / 5.4B active
MoE total / active
Context
1,048,576 tokens
native
Modalities
text, image text
input → output
Benchmarks
#1/138
better than 99% of all scored models
Size class
#1/5
better than 80% of similar-size models
License
Apache License 2.0
About
Inkling 2 Mini is a general-purpose multimodal open-weights model from Thinking Machines Lab, Inc. (release: November 20, 2026, license: Apache 2.0, model family Inkling, alongside Inkling 2 Small). This record refers to the official weights distribution on Hugging Face, thinkingmachines/Inkling-2-Mini; the checkpoint ships as a single ~20 GB file in mixed NVFP4 and MXFP8 precisions. Unsloth publishes quantized GGUF derivatives in the unsloth/Inkling-2-Mini-GGUF repository (Dynamic 3.0 / imatrix UD-* quants from UD-IQ1_S at 7.76 GB up to UD-Q8_K_XL at 22.87 GB, with a per-quant analysis of Mean KLD vs UD-Q8_K_XL and same-top-token rate, mainline llama.cpp uploads under Mainline_Support, and Unsloth Desktop support; the card of that GGUF repo was the source document of this ingest).
Model
Inkling 2 Mini accepts text and image inputs and generates text outputs. It is intended for English and other languages and across multiple coding languages, and for developers building AI-powered applications: agentic and tool-use systems, coding assistants, chatbots, and retrieval-augmented generation, plus general-purpose conversational use and instruction following. It supports a context window of up to 1M tokens. Images are encoded by a vision transformer (inkling2_vision; 40px–4096px per dimension recommended) and projected into a shared hidden space processed jointly by the decoder. (Model card sections 1 and 3)
Looped architecture
The architecture type is unusual: a multimodal autoregressive looped transformer — a 24-layer decoder-only transformer whose middle 12 layers run three times with shared weights per token (48 layer applications from 24 layers of weights; model_type: inkling2_loop, class Inkling2LoopForConditionalGeneration in the config.json). The feed-forward path is a sparse Mixture-of-Experts with 128 experts, routing each token to 8 of 128 experts. Attention is a hybrid of local and global layers; the config records GQA with 24 query heads / 8 KV heads and head_dim 128 at hidden_size 2304 and vocab_size 201024. Parameters: 26.2B total, 3.4B active nominally — each extra pass through the looped layers adds about 1.0B effective active parameters, giving 5.4B effective active parameters per token across three passes. Locally, the model runs on 1x NVIDIA DGX Spark, Apple silicon with ≥32 GB of memory, or GPUs with ≥32 GB VRAM. (Model card section 3; config.json)
Looping is the first release of this design by Thinking Machines: it lets the model spend more computation per token without reasoning in visible tokens. The chain of thought has very little controllability, and in the monitorability study Inkling 2 Mini was unable to perform covert side tasks in realistic agentic environments; the company's blog post describes the monitorability experiments. (Model card section 6)
Training and alignment
Training used a broad mix of content types — text, images, video — from publicly available sources, third-party acquired data, internally generated and synthetic/augmented data, with deduplication and filtering curation. As a pre-release risk mitigation there was an extra alignment-training phase using on-policy distillation, similar to Inkling 2 Small. Safety evaluations covered everyday interaction (sycophancy, suicide/self-harm and eating-disorder requests, refusal balance), CBRN/cyber uplift, agentic loss-of-control (deception, sabotage) and monitorability; the card concludes the model stays materially below frontier capabilities and gives no material uplift beyond existing open-weights models. (Model card sections 4 and 6)
Evaluations
The model card benchmarks Inkling 2 Mini against Gemma 4 31B, Muse Glimmer, Qwen3.8-27B and Nemotron 3.5 Super VL. Highlights: SWEBench Verified (bash only) 79.2%, SWEBench Pro v2 (Hard) 77.9%, SWEBench Multilingual 73.7%, Terminal Bench v2.1 (pi v0.83.0) 71.6%, Vibe Code Bench v1.1 77.9%, BrowseComp 79.0%, DeepSearchQA 85.3%, MCP Atlas 82.7%; third-party indices: AA Index v5.0 20.1 (tie with Qwen3.8-27B), Vals Index v2.2 46.5%. Vision: MMMU Pro 74.9%, OmniDocBench v1.6 92.8%, ScreenSpot Pro 76.5% (87.7% with Python). Stated limitations include weaker factual recall than Inkling 2 Small, repetitive reasoning loops, less reliable agentic tool behavior (including possible destructive actions without confirmation), weaker non-English output, reasoning effort minimum not fully disabling reasoning, and higher per-token compute from the 3x looping. (Model card sections 5–7)
Access
Official weights are on Hugging Face (thinkingmachines/Inkling-2-Mini) and via the Tinker fine-tuning platform (API access, Tinker playground), with the Tinker Cookbook (github.com/thinking-machines-lab/tinker-cookbook) and the tml-renderers package. GGUF inference is documented by Unsloth in its Inkling-2-Mini guide; the quants sit in the Unsloth Dynamic 3 Quants collection (Dynamic v3.1/v3.0, best on Terminal Bench, DeepSWE and more at 512K context). (Model card)
Training Data Broad mix of content types: text, images and video, drawn from publicly available sources, third-party acquired data, internally generated and synthetic/augmented data. Curation includes deduplication and filtering (Model Card §4); final alignment phase used on-policy distillation (§6). Documentation: https://thinkingmachines.ai/training-data-documentation/
Benchmark Scores
| Benchmark | Score | Date | Standing |
|---|---|---|---|
|
Artificial Analysis Intelligence Index v5
composite
|
20.10%
|
20.11.2026 |
#6/8
better than 14%
|
|
FrontierCode 2.0
coding_agent
|
15.10%
|
20.11.2026 |
#2/4
better than 67%
|
|
DeepSWE 1.1
coding_agent
|
48.20%
|
20.11.2026 |
#24/38
better than 38%
|
|
Terminal-Bench 4.0
coding_agent
|
5.60%
|
20.11.2026 |
#11/13
better than 8%
|
|
Terminal Bench 2.1
coding_agent
|
71.60
|
20.11.2026 |
#24/47
better than 50%
|
|
SWE-bench Pro v2.0 Hard
coding_agent
|
77.90%
|
20.11.2026 |
#2/8
better than 86%
|
|
SWE-bench Multilingual
coding_agent
|
73.70%
|
20.11.2026 |
#22/51
better than 58%
|
|
SWE-bench Verified
coding_agent
|
79.20%
|
20.11.2026 |
#12/62
better than 82%
|
|
BrowseComp
general_agent
|
79.00%
|
20.11.2026 |
#19/55
better than 67%
|
|
DeepSearch QA
general_agent
|
85.30%
|
20.11.2026 |
#10/19
better than 50%
|
|
MCP-Atlas
general_agent
|
82.70%
|
20.11.2026 |
#6/57
better than 91%
|
|
AA-Briefcase v1.1
general_agent
|
1263.00
|
20.11.2026 |
#5/6
better than 20%
|
|
GDPval-AA v2.1
general_agent
|
1319.00
|
20.11.2026 |
#5/6
better than 20%
|
|
JobBench
general_agent
|
49.60%
|
20.11.2026 |
#14/25
better than 46%
|
|
Automation Bench-AA
agentic
|
41.80%
|
20.11.2026 |
#5/6
better than 20%
|
|
EnterpriseOps-Gym
general_agent
|
42.00%
|
20.11.2026 |
#2/5
better than 75%
|
|
GDP.pdf
long_context
|
16.80%
|
20.11.2026 |
#7/14
better than 54%
|
|
PerceptionBench
vision_language
|
57.50%
|
20.11.2026 |
#2/6
better than 80%
|
|
MMMU-Pro
vision_language
|
74.90%
|
20.11.2026 |
#21/39
better than 47%
|
|
ScreenSpot Pro
general_agent
|
76.50
|
20.11.2026 |
#2/14
better than 92%
|
|
Chartography
multimodal_agent
|
22.10
|
20.11.2026 |
#8/10
better than 22%
|
|
ChartoGraphy w/ python
vision_language
|
72.50%
|
20.11.2026 |
#3/5
better than 50%
|
|
ZEROBench
vision_language
|
17.00
|
20.11.2026 |
#9/19
better than 56%
|
|
ZeroBench_main w/ python (pass@5)
vision_language
|
34.00%
|
20.11.2026 |
#3/4
better than 33%
|
|
MuirBench
video_understanding
|
77.50%
|
20.11.2026 |
#3/5
better than 50%
|
|
HLE-Diamond (no tools)
stem_reasoning
|
15.20%
|
20.11.2026 |
#6/7
better than 17%
|
|
MathArena Apex 2025
stem_reasoning
|
34.40%
|
20.11.2026 |
#9/11
better than 20%
|
|
LongBench v2
long_context
|
66.90
|
20.11.2026 |
#3/28
better than 93%
|
|
OmniScience Non-Hallucination
knowledge
|
63.40%
|
20.11.2026 |
#7/13
better than 50%
|
|
FORTRESS Benign
safety
|
96.40%
|
20.11.2026 |
#2/8
better than 86%
|
|
StrongREJECT v2 (adaptive)
safety
|
97.80%
|
20.11.2026 |
#5/8
better than 43%
|
|
Vals Index (v2.2)
composite
|
46.50%
|
20.11.2026 |
#1/1
only tracked model
|
|
Vibe Code Bench v1.1
coding
|
77.90%
|
20.11.2026 |
#1/2
better than 100%
|
|
PMPP-Hard (AUC, 0-10h)
coding
|
455.82
|
20.11.2026 |
#1/2
better than 100%
|
|
Vals Finance Index
finance
|
50.10%
|
20.11.2026 |
#1/1
only tracked model
|
|
AA-AnalystAgent
general_agent
|
46.30%
|
20.11.2026 |
#1/2
better than 100%
|
|
Diligence Bench
finance
|
48.60%
|
20.11.2026 |
#1/3
better than 100%
|
|
OmniDocBench v1.6
vision
|
92.80%
|
20.11.2026 |
#1/3
better than 100%
|
|
ScreenSpot Pro (with Python)
vision
|
87.70%
|
20.11.2026 |
#1/2
better than 100%
|
|
BVB
vision
|
68.40%
|
20.11.2026 |
#2/3
better than 50%
|
|
CritPt v2 (Full)
reasoning
|
31.40%
|
20.11.2026 |
#1/2
better than 100%
|
|
CritPt v2 (Hard)
reasoning
|
17.00%
|
20.11.2026 |
#1/3
better than 100%
|
|
ARC-AGI (v1)
reasoning
|
84.50%
|
20.11.2026 |
#2/2
better than 0%
|
|
ARC-AGI (v2)
reasoning
|
49.30%
|
20.11.2026 |
#1/2
better than 100%
|
|
GraphWalks (BFS, 256K)
long_context
|
69.70%
|
20.11.2026 |
#1/2
better than 100%
|
|
MRCR v2 (8-needle, 256K-512K)
long_context
|
46.40%
|
20.11.2026 |
#1/2
better than 100%
|
|
AA-Omniscience v1.1 (score)
knowledge
|
-8.60
|
20.11.2026 |
#2/3
better than 50%
|
|
ComplexConstraints v1.1
agentic
|
33.40%
|
20.11.2026 |
#2/3
better than 50%
|
|
Misbehavior (Internal)
safety
|
9.80%
|
20.11.2026 |
#3/3
better than 0%
|
|
Covert misbehavior (Internal)
safety
|
8.30%
|
20.11.2026 |
#3/3
better than 0%
|
|
Destructive actions (Internal)
safety
|
10.20%
|
20.11.2026 |
#3/3
better than 0%
|
What is a good Artificial Analysis Intelligence Index v5 score? What is a good FrontierCode 2.0 score? What is a good DeepSWE 1.1 score?
Hardware Requirements
Estimated VRAM per quantization at 4,096 tokens context — MoE: all experts must fit in VRAM, only 5.4B are active per token
| Quantization | Weights | KV Cache | Total VRAM | On your hardware |
|---|---|---|---|---|
| FP16 — 16-bit | 52.4 GB | 0.8 GB | 55.9 GB | Which GPUs can run it? Calculate on your GPU |
| INT8 — 8-bit | 26.2 GB | 0.4 GB | 27.9 GB | Which GPUs can run it? Calculate on your GPU |
| INT4 — 4-bit | 13.1 GB | 0.2 GB | 14.0 GB | Which GPUs can run it? Calculate on your GPU |
| FP4 — 4-bit | 13.1 GB | 0.2 GB | 14.0 GB | Which GPUs can run it? Calculate on your GPU |
Consumer GPUs that fit this model with a single card (INT4):
Architecture
decoder-only transformer
- Parameters
- 26.2B
- Active params
- 5.4B
- Layers
- 24 24× Full Attention
- Attention
- GQA 24:8 Grouped Query Attention
- Experts
- 128 top-8
- Hidden
- 2304
- Head dim
- 128
- Vocabulary
- 201K
- Context
- 1M
Source: Hugging Face config.json · Inkling2LoopForConditionalGeneration · model repo
Training Pipeline
-
1
pretraining
General multimodal pretraining
The card (section 4) describes broad multimodal training over a variety of content types including text, images, and video, with training data drawn from publicly available sources, acquired from third-parties, generated internally, or synthetically generated or augmented; publicly available data includes content from the public internet and publicly accessible repositories. Data curation included cleaning, processing, and modifying datasets, with steps varying by data type and potentially including deduplication and filtering to remove junk or other low-quality data, or to advance safety or other objectives. No training hyperparameters, duration, or cost are stated on the card for this phase.
-
2
rl
Alignment phase via on-policy distillation
Stage-type mapping rationale: on-policy distillation is a reinforcement-learning-family method (on-policy student rollouts scored by a teacher with reward-style per-token objectives) and the card calls it an alignment-training mitigation applied before release, similar to Inkling 2 Small (card section 6). Beyond this phase the card names no further post-training recipe or hyperparameters.
Linked Resources
How to Run Inkling-2-Mini Guide
https://unsloth.ai/docs/models/inkling-2-mini
Unsloth Dynamic 3.0 GGUFs
https://unsloth.ai/docs/basics/unsloth-dynamic-v3.0-gguf
unslothai/unsloth
https://github.com/unslothai/unsloth
Tinker Playground
https://tinker.thinkingmachines.ai/playground
thinking-machines-lab/tinker-cookbook
https://github.com/thinking-machines-lab/tinker-cookbook
Model Acceptable Use Policy
https://thinkingmachines.ai/model-acceptable-use-policy/
tml-renderers
https://pypi.org/project/tml-renderers/
Inkling 2 Mini blog post
https://thinkingmachines.ai/news/inkling-2-mini/
Training Data Documentation
https://thinkingmachines.ai/training-data-documentation/
Unsloth Dynamic 3 Quants
https://huggingface.co/collections/unsloth/unsloth-dynamic-3-quants
Unsloth Discord
https://discord.gg/unsloth
unsloth/inkling-2-mini (HF model page, documentation button)
https://huggingface.co/unsloth/inkling-2-mini
Mainline_Support folder (mainline llama.cpp uploads)
https://huggingface.co/unsloth/Inkling-2-Mini-GGUF/tree/main/Mainline_Support
unsloth/Inkling-2-Mini-GGUF (Unsloth Dynamic 3.0 GGUF quants)
https://huggingface.co/unsloth/Inkling-2-Mini-GGUF
No comments yet
Be the first to share your thoughts!